Principal Machine Learning Engineer

HubSpot•Cambridge, MA

7d•$285,800 - $457,300•Hybrid

About The Position

HubSpot is building the foundational AI infrastructure that powers our next generation of agentic product experiences. We’re looking for a Principal Machine Learning Engineer to help shape the base layer that product teams build on: agent runtime, evaluation systems, quality signals, model optimization, fine-tuning, and model routing. This team owns Breeze Engine, the platform behind Breeze Assistant and other AI-powered product experiences, as well as Aviator, HubSpot’s internal framework for building and running agents. This is not a feature-team role it is a high-scope infrastructure role focused on building systems that create leverage across the company.

Requirements

Deep experience with production ML, LLM, or AI infrastructure.
Built systems at scale where reliability, latency, quality, and cost matter.
Thinks in platforms, not one-off solutions.
Can work across infrastructure, product, backend, and ML teams.
Experience with areas like eval frameworks, model serving, fine-tuning, signal extraction, experimentation, feedback loops, or model optimization.
Excited by ambiguous, foundational problems where many of the answers still need to be invented.
Can influence senior engineers and leaders through technical credibility and clear judgment.

Responsibilities

Build infrastructure for agent runtime, evaluation, quality measurement, and model optimization.
Create tooling that helps teams understand where their evals are strong, where coverage is missing, and where users are asking questions the system is not yet prepared to handle.
Design signal pipelines that surface user frustration, agent failure states, and quality issues before they show up as CSAT drops.
Help define a HubSpot-specific AI benchmark for evaluating model performance on real HubSpot workloads.
Build systems to evaluate new models across agents, improving product quality, reducing cost, and supporting future model routing.
Make fine-tuning and task-specific optimization more repeatable for HubSpot’s AI use cases.